Chemical incident emergency response method and system under real-time monitoring

Abstract

The application provides a chemical accident emergency response method and system under real-time monitoring, relates to the technical field of emergency response, time alignment is carried out through real-time collection of multi-source heterogeneous data to generate multi-source synchronous data; multi-modal features of the multi-source synchronous data are extracted for chemical accident identification; accident scene parameters are called, accident response analysis is carried out based on the accident identification result, an emergency response scheme is formulated and resource scheduling analysis is carried out, a resource scheduling network diagram is constructed; an accident monitoring and early warning signal is carried out according to the resource scheduling network diagram combined with the multi-source synchronous data, real-time decision instructions are generated, and the emergency response scheme is pushed to an emergency command center. The application solves the technical problems that the existing technology lacks multi-source heterogeneous data fusion and real-time linkage mechanism, leading to low chemical accident identification accuracy and lagged response process, and achieves the technical effects of improving chemical accident identification accuracy and emergency response speed.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of emergency response, in particular to a chemical accident emergency response method and system under real-time monitoring. BACKGROUND

[0002] In the process of chemical production, storage and transportation, and disposal, it is of great significance to build an efficient and intelligent chemical accident emergency response mechanism to protect people's life and property safety and ecological environment. At present, the chemical accident emergency response mainly relies on traditional sensing devices and preset response plans, and the accident is judged based on single sensor data or manual alarm signals, and the resources are allocated by manual command to execute the response task. However, in practical application, due to the lack of collaborative collection and real-time fusion of various types of data such as sensor data and video monitoring data, the comprehensive sensing ability of the accident scene is insufficient, which affects the accurate identification of the accident type and risk level. At the same time, the existing method relies on fixed process and static plan, and lacks the mechanism of real-time scene linkage, so the connection between the links from accident monitoring to emergency response is not close, the response process is lagging behind, and it cannot meet the demand of fast and accurate emergency response. SUMMARY

[0003] The present application provides a chemical accident emergency response method and system under real-time monitoring, which solves the technical problems of low chemical accident identification accuracy and lagging response process caused by the lack of multi-source heterogeneous data fusion and real-time linkage mechanism in the prior art, and achieves the technical effect of improving the chemical accident identification accuracy and emergency response speed.

[0004] In view of the above problems, on the one hand, the present application provides a chemical accident emergency response method under real-time monitoring, which comprises: collecting multi-source heterogeneous data in real time, time aligning based on the multi-source heterogeneous data, and generating multi-source synchronous data; performing multi-modal deep learning according to the multi-source synchronous data, extracting multi-modal features, identifying chemical accidents according to the multi-modal features, and obtaining accident identification results; calling accident scene parameters, performing accident response analysis based on the accident identification results, formulating an emergency response scheme, performing resource scheduling analysis according to the emergency response scheme, and constructing a resource scheduling network graph; monitoring the accident according to the resource scheduling network graph combined with the multi-source synchronous data, triggering an early warning signal according to the monitoring result, generating a real-time decision instruction, and pushing the real-time decision instruction combined with the emergency response scheme to an emergency command center to respond to the chemical accident.

[0005] In another aspect, the application also provides a chemical accident emergency response system under real-time monitoring, comprising: a data acquisition module for acquiring multi-source heterogeneous data in real time, performing time alignment based on the multi-source heterogeneous data, and generating multi-source synchronous data; an accident identification module for performing multi-modal deep learning based on the multi-source synchronous data, extracting multi-modal features, identifying a chemical accident according to the multi-modal features, and obtaining an accident identification result; a response analysis module for calling accident scene parameters, performing accident response analysis based on the accident identification result, formulating an emergency response scheme, performing resource scheduling analysis of the emergency response scheme, and constructing a resource scheduling network graph; and an emergency response module for monitoring the accident according to the resource scheduling network graph in combination with the multi-source synchronous data, triggering an early warning signal according to a monitoring result, generating a real-time decision instruction, and pushing the real-time decision instruction to an emergency command center in combination with the emergency response scheme to perform emergency response to the chemical accident.

[0006] The one or more technical solutions provided in the application have at least the following beneficial effects:

[0007] By acquiring multi-source heterogeneous data in real time and performing time alignment to generate multi-source synchronous data, the data from different sources is ensured to have synchronicity and consistency, providing a comprehensive and accurate data basis for subsequent accident analysis and decision-making, and avoiding one-sidedness in judgment caused by a single data source. Multi-modal deep learning is performed based on the multi-source synchronous data to extract multi-modal features and identify a chemical accident, which utilizes deep learning technology to extract features from multi-modal data, can more accurately identify key information of a chemical accident, improves the accuracy of accident identification, and provides a basis for taking correct emergency measures. By calling accident scene parameters to perform accident response analysis, formulating an emergency response scheme, and performing resource scheduling analysis to construct a resource scheduling network graph, it is helpful to more efficiently allocate emergency resources, ensure that resources can quickly and reasonably reach the accident site as needed after an accident occurs, improve the efficiency of emergency response, and reduce the loss caused by the accident. According to the resource scheduling network graph in combination with the multi-source synchronous data, the accident is monitored, an early warning signal is triggered, a real-time decision instruction is generated, and the real-time decision instruction is pushed to the emergency command center, forming a dynamic closed-loop monitoring and response mechanism. By continuously monitoring the accident situation and combining it with the resource scheduling network graph, changes and abnormalities in the accident can be discovered in time, an early warning signal is triggered, decision makers can quickly generate and adjust decision instructions according to real-time information, and emergency response measures can be timely and effectively implemented, enhancing the flexibility and real-time performance in dealing with accidents.

[0008] In summary, this application achieves accurate identification and rapid assessment of chemical accidents through real-time acquisition and fusion of multi-source heterogeneous data, combined with multimodal deep learning technology. Simultaneously, a resource scheduling network diagram constructed based on accident scenario and resource scheduling analysis optimizes resource allocation and scheduling efficiency in emergency response. Finally, through real-time monitoring, early warning, and rapid delivery of decision-making instructions, a closed-loop emergency response mechanism is formed, enabling continuous linkage and intelligent adjustment capabilities. Overall, this solution effectively improves the identification accuracy, response speed, and collaborative handling capabilities of chemical accidents, providing technical support and intelligent assurance for emergency management in high-risk scenarios.

[0009] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0010] Figure 1 This is a flowchart illustrating a chemical accident emergency response method under real-time monitoring, as provided in an embodiment of this application.

[0011] Figure 2 This is a schematic diagram of the structure of a real-time monitoring-based emergency response system for chemical accidents provided in an embodiment of this application.

[0012] Explanation of reference numerals in the attached diagram: Data acquisition module 10, accident identification module 20, response analysis module 30, emergency response module 40. Detailed Implementation

[0013] This application provides a method and system for emergency response to chemical accidents under real-time monitoring, which solves the technical problems of low accuracy in chemical accident identification and delayed response process caused by the lack of multi-source heterogeneous data fusion and real-time linkage mechanism in the prior art, and achieves the technical effect of improving the accuracy of chemical accident identification and the speed of emergency response.

[0014] Example 1, as Figure 1 As shown in the embodiments of this application, a method for emergency response to chemical accidents under real-time monitoring is provided, the method comprising:

[0015] Step S100: Collect multi-source heterogeneous data in real time, perform time alignment based on the multi-source heterogeneous data, and generate multi-source synchronized data.

[0016] Specifically, the multi-source heterogeneous data includes hazardous chemical risk monitoring and early warning system data, chemical basic information in enterprise basic information database, video monitoring network data, etc. Among them, the hazardous chemical risk monitoring and early warning system data includes real-time monitoring data uploaded from toxic gas sensors, environmental parameter collectors (temperature, humidity, wind speed, etc.), dangerous goods storage tank liquid level and pressure sensors and other equipment, usually in structured or semi-structured format, with strong time sequence. The enterprise basic information database contains static structured data such as registration information, storage location, dangerous goods category, emergency contact, emergency material configuration, and historical safety record of hazardous chemical using units. Video monitoring network data includes live video image stream covering key areas inside and outside the factory, such as storage tank area, loading and unloading area, transportation route, etc., and the data is unstructured video frame stream or image sequence.

[0017] Connect with hazardous chemical risk monitoring and early warning system, enterprise basic information database, and video monitoring network through data interface, real-time collect these multi-source heterogeneous data, then standardize structured data into unified format (such as JSON, CSV), frame processing for video data, extract key frames (such as 1 frame per second) and record corresponding time stamp. Introduce sliding time window (such as 5 seconds), map multi-source heterogeneous data to unified time anchor point according to time stamp, if there is missing data of a certain type, use linear interpolation or historical average value for compensation, so that all data are time-aligned. All data complete time alignment to form multi-source synchronous data structure. There are corresponding data from different data sources at each time point in multi-source synchronous data, which is convenient for subsequent comprehensive analysis and processing.

[0018] This step generates multi-source synchronous data by real-time collecting multi-source heterogeneous data from hazardous chemical risk monitoring and early warning system, enterprise basic information database, and video monitoring network, and time aligning, which provides a comprehensive, accurate and time-consistent data basis for subsequent chemical accident identification and emergency response, so that the subsequent emergency response process can consider various information of the accident scene, so as to more accurately judge the accident situation and improve the efficiency and accuracy of emergency response.

[0019] Step S200: According to the multi-source synchronous data, multi-modal deep learning is performed to extract multi-modal features, and chemical accident identification is performed according to the multi-modal features to obtain accident identification results.

[0020] Specifically, multi-modal deep learning is a method of using deep learning algorithm to jointly analyze and learn multi-modal (such as image, time series signal, text) data, which can automatically mine the correlation and deep features between different modal data to better understand complex scenes.

[0021] A multi-channel deep learning model is constructed, such as a fusion architecture based on a convolutional neural network for processing image video data, a recurrent neural network or a long short-term memory network for processing time series sensor data, and a natural language processing model for processing text data. The fusion architecture is trained using a large amount of labeled and unlabeled chemical accident related data, learns the feature representation of different modal data, and fuses these features through attention mechanisms or weighted averaging, etc. For example, a multi-modal deep learning model is constructed using deep learning frameworks such as TensorFlow or PyTorch, three-dimensional convolution analysis is performed on video stream data to extract spatio-temporal features, long short-term memory network analysis is performed on sensor time series signals to extract time series features, and structured text data is processed based on word embedding and attention mechanism to extract semantic keyword features. Then, these features are fused by weighted fusion (weights can be automatically learned by the model or manually set according to the importance of the features) to construct multi-modal features.

[0022] The multi-source synchronous data generated in step S100 is input into the multi-channel deep learning model to extract feature vectors with representativeness and discrimination, and to obtain multi-modal features. These features can comprehensively reflect various key information of the current accident, such as flame color, size, shape features extracted from video, gas leakage rate, concentration peak features extracted from sensor data, accident type description, emergency keyword features extracted from text, etc. According to the constructed multi-modal features, a classification algorithm (such as support vector machine, decision tree, etc.) or a fully connected classification layer is added at the last layer of the multi-channel deep learning model to identify the chemical accident, and to output an accident identification result containing the current accident type (such as fire, leakage, explosion, etc.) and the accident severity (such as minor, general, major, etc.).

[0023] This step realizes multi-modal information fusion intelligent perception, effectively improves the identification accuracy and judgment speed of the accident scene, provides a key basis for subsequent development of a reasonable emergency response scheme, and improves the pertinence and timeliness of the emergency response.

[0024] Step S300: Retrieve accident scene parameters, perform accident response analysis based on the accident identification result, develop an emergency response scheme, execute the emergency response scheme for resource scheduling analysis, and construct a resource scheduling network diagram.

[0025] Specifically, the accident scene parameters are parameters describing various states and conditions of the accident site, including the location, time, types and quantities of chemicals, meteorological conditions (wind direction, wind speed, temperature, etc.), surrounding environment (buildings, population density, traffic conditions, etc.) of the accident. For example, the accident occurs in a large chemical warehouse, which stores various dangerous chemicals, the wind direction is northeast, the wind speed is 5 m / s, and there are residential areas within a range of 500 meters.

[0026] The current accident scene parameters are retrieved from data sources such as the enterprise's safety management information system, geographic information system (GIS), data interface of the meteorological department, etc. According to the accident identification result obtained in step S200, the accident response analysis is performed using the pre-established accident response model (such as modeling based on reinforcement learning), and the optimal emergency response scheme is formulated in the state-action space, including task allocation, personnel evacuation path, resource calling, etc. Subsequently, the rescue resource distribution, traffic information and team position are extracted from the geographic information system, and the resource scheduling network graph is constructed using graph theory tools (such as Dijkstra or A-STAR algorithm), with nodes representing sites or resources and edges representing paths. The resource scheduling network graph displays the scheduling path, nodes and priority of emergency rescue resources in a graphical manner, intuitively presenting the flow process and time arrangement of resources from the supply point to the accident site, providing decision support for emergency command personnel to conduct resource scheduling.

[0027] Step S400: According to the resource scheduling network graph and the multi-source synchronous data, the accident is monitored, the pre-warning signal is triggered according to the monitoring result, the real-time decision instruction is generated, and the real-time decision instruction is combined with the emergency response scheme to push to the emergency command center for emergency response to the chemical accident.

[0028] Specifically, during the emergency response process, the resource scheduling network graph constructed in step S300 and the multi-source synchronous data constructed in step S100 are combined to continuously monitor the accident site and surrounding environment in real time, track the development dynamics of the accident, including the change of chemical leakage and diffusion range, the trend of fire spread, the change of on-site meteorological conditions, the personnel casualty situation, etc. The real-time monitoring result is compared with the preset monitoring index threshold (such as gas concentration threshold, flame height threshold, etc.), when the real-time monitoring result exceeds the threshold, the corresponding pre-warning signal is triggered using the pre-warning rule engine, and the real-time decision instruction for guiding the adjustment and optimization of on-site rescue action is generated through the strategy engine, the real-time decision instruction is integrated with the emergency response scheme, and the real-time decision instruction is pushed to the emergency command center through the emergency command communication system (such as satellite communication, trunking radio, emergency command dispatching software, etc.) for manual confirmation or automatic execution.

[0029] The step realizes dynamic closed-loop control in the accident process, provides real-time decision support for emergency command personnel, ensures that the emergency response measures can be flexibly adjusted according to the accident development, and enhances the flexibility and adaptability of the emergency response.

[0030] Further, the step S100 comprises:

[0031] Step S110: Real-time acquisition is performed through a multi-source data interface to obtain multi-source heterogeneous data.

[0032] Step S120: A time window length is dynamically set to construct a time window, and the multi-source heterogeneous data is aligned according to timestamps by using the time window to generate an aligned data set.

[0033] Step S130: Time stamp offset analysis is performed according to the aligned data set to generate a plurality of offsets.

[0034] Step S140: It is judged whether the plurality of offsets exceeds the time window length, and if any offset in the plurality of offsets exceeds the time window length, a data compensation instruction is triggered.

[0035] Step S150: The aligned data set is compensated for missing data by using the data compensation instruction to obtain the multi-source heterogeneous data.

[0036] Specifically, by configuring an industrial bus interface, a network API interface, a database connector or a message queue mechanism, sensor monitoring data, enterprise static information and video images and other heterogeneous types of data in a plurality of data sources such as a dangerous chemical risk monitoring and early warning system, an enterprise basic information database and a video monitoring network are accessed and collected. A uniform timestamp is added to each data item during the collection process, realizing real-time synchronous access of multi-source data.

[0037] According to the uploading delay characteristics and frequency distribution of various data sources, the time window length is dynamically adjusted, for example, to 5 seconds or 10 seconds, and the multi-source data is time-matched in each time window to form a combined structure of video frames, sensor data and enterprise attribute information under the same window, and to construct an aligned data set.

[0038] For each data source in the aligned data set, the time difference between the actual timestamp of each data item and the center anchor point of its corresponding time window is calculated. These time differences are the timestamp offsets, which are used to measure the degree of time deviation of different data sources. For example, the recording time of a certain sensor data is 10:31:03, and the corresponding window anchor point time is 10:31:00, so the offset is +3 seconds.

[0039] The timestamp offset of each data source is compared with the length of the time window, and it is determined whether the offset exceeds the length of the time window. If any offset in the multiple offsets exceeds the length of the time window, it is considered that the data item has been invalidated or lost, and a data compensation instruction is triggered. The instruction contains information such as the data source identifier to be compensated and the time range of the missing data, and is used to guide subsequent data compensation operations.

[0040] According to the data compensation instruction, different compensation strategies are selected for different data types, including linear interpolation, adjacent time replacement, historical data filling, or model-based predicted value replacement, to complete the missing or invalid data. For example, for hazardous chemical sensor data, a linear interpolation method can be used to estimate the data value at the missing time point according to the existing sensor data points in the adjacent time window; for video frame loss, the previous frame can be used for replacement; for enterprise database field loss, the default or latest value can be used; or if the data source supports data retransmission function, the data source can be requested to resend the data in the specified time range. The compensated data is replaced or added to the original aligned data set to form a multi-source synchronized data set with complete structure and good time consistency.

[0041] The above steps realize efficient access, dynamic timing coordination and robust compensation fusion of multi-source heterogeneous data, and construct a multi-source synchronized data set with high consistency and low delay. This data set serves as the basis for subsequent accident identification and response analysis, effectively solving the problems of data island, synchronization inaccuracy and serious delay in traditional systems, and fundamentally improving the timeliness and accuracy of chemical accident intelligent perception.

[0042] Further, step S120 comprises:

[0043] Step S121: retrieve historical transmission delay statistical values of multiple data sources in the multi-source data interface, and initialize the reference length of the time window according to the historical transmission delay statistical values.

[0044] Step S122: real-time monitor the transmission jitter rate of the multiple data sources, and generate an expansion parameter when the transmission jitter rate exceeds a preset threshold.

[0045] Step S123: dynamically expand the reference length of the time window according to the expansion parameter, and set the length of the time window.

[0046] Specifically, by calling the historical data interface, the average transmission delay and fluctuation interval of various data sources (such as video stream, sensor sampling signal, database record) in a period of time are obtained. The initial reference length of the time window is determined according to the historical transmission delay statistical value of each data source. Usually, the maximum value, average value or other representative statistical quantity of the historical transmission delay of each data source is selected as the reference of the reference length of the time window. For example, the average delay of sensor data is 1.8 seconds, and the average delay of video frame data is 2.5 seconds. The initial length of the reference window can be set to 3 seconds to accommodate the normal delay range of the data.

[0047] The transmission jitter rate refers to the fluctuation degree of the transmission delay of the data source in the process of transmitting data. It reflects the stability of data transmission. The transmission jitter standard deviation or coefficient of variation is calculated based on the sliding window statistical method as the transmission jitter rate. If the transmission jitter rate exceeds the preset threshold, it is considered that the current data transmission fluctuates violently, and the time window needs to be adjusted. At this time, an expansion parameter is automatically generated. This expansion parameter is a proportion value, which represents the proportion of the length that the time window needs to be expanded on the initial value. For example, the expansion parameter can be set to 20%, which means that the time window length needs to be expanded by 20% based on the reference length.

[0048] The length of the time window is increased by the proportion specified by the expansion parameter. For example, when the reference length is 3 seconds and the expansion parameter is 1.5, the final length of the time window is set to 4.5 seconds. This expansion method can automatically tolerate a larger range of data delay during data abnormal fluctuation, improving the robustness of time alignment.

[0049] Through the above steps, adaptive analysis and window adjustment of the transmission characteristics of multi-source heterogeneous data can be realized, ensuring that the time tolerance in the data alignment process is reasonably dynamic, and effectively enhancing the accuracy of data fusion.

[0050] Further, step S200 comprises:

[0051] Step S210: inputting the multi-source synchronous data set into a multi-modal deep learning network for multi-modal analysis to obtain multi-modal data, wherein the multi-modal data includes video stream data, sensor time series signal and structured text data.

[0052] Step S220: performing three-dimensional convolution analysis based on the video stream data to extract spatio-temporal features.

[0053] Step S230: performing bidirectional analysis based on the sensor time series signal to extract time series features.

[0054] Step S240: performing natural language processing based on the structured text data to extract semantic keyword features.

[0055] Step S250: The spatio-temporal features, the time series features, and the semantic keyword features are weightedly fused to construct the multi-modal features.

[0056] Specifically, the multi-source synchronous data set is input into a multi-modal deep learning network for analysis to obtain multi-modal data. The multi-modal data includes image sequences (video stream data) from video monitoring, time series signals (sensor time series signals) obtained by a dangerous chemical risk monitoring and early warning system, and structured text data in an enterprise basic information database. After time alignment processing, the above data has time synchronization characteristics and can be input into a multi-modal learning channel in parallel.

[0057] In the video stream data processing channel, a three-dimensional convolutional neural network is used for spatio-temporal analysis of images to extract the spatial position and dynamic change characteristics of accident precursors (such as flame flashing and smoke diffusion) in consecutive pictures. For example, if the camera monitors rapid diffusion of smoke in the storage tank area, the three-dimensional convolutional neural network can determine the spreading speed and direction.

[0058] In the sensor data processing channel, a bidirectional long short-term memory network is used to analyze sensor time series signals to extract time-dependent relationships and trend characteristics between variables, such as abnormal mutations or regular fluctuations in sensor signals such as temperature, pressure, and concentration. For example, for gas leakage sensor data of a certain storage tank, the bidirectional long short-term memory network can consider both past and future data points to capture the change pattern of gas concentration over time, such as rapid concentration rise in the early stage of leakage, concentration fluctuation in the stable stage of leakage, and possible concentration mutation, which helps to predict the potential development trend of the accident.

[0059] In the text data processing channel, a natural language processing model such as BERT is used to perform semantic analysis on structured text to extract representative keywords and semantic labels, such as “corrosive liquid”, “overstocking”, “explosion threshold”, and other descriptions closely related to accidents.

[0060] The spatio-temporal features, time series features, and semantic keyword features output by the above three channels are weightedly fused to construct a unified multi-modal feature vector. The weighting strategy can adjust the fusion weights through attention mechanisms or projection fusion algorithms based on the performance differences of each data source in the training stage to form a comprehensive perception expression of the chemical accident state.

[0061] Through the above multi-modal deep learning processing, not only the spatio-temporal and semantic information of multi-source data is effectively integrated, but also a high-reliability input basis is provided for the subsequent steps of accident identification and response strategy formulation, significantly improving the intelligent identification capability of complex and hidden accident situations.

[0062] Further, step S300 includes:

[0063] Step S310: vectorization based on the accident scene parameters, definition of state space, emergency action matching based on the accident identification results, definition of action space.

[0064] Step S320: reinforcement learning of the state space combined with the action space to formulate an emergency response scheme.

[0065] Step S330: resource scheduling analysis according to the emergency response scheme to obtain rescue resource distribution data, traffic network information, and real-time location of rescue teams.

[0066] Step S340: construction of multiple nodes based on the rescue resource distribution data, construction of multiple edges of the nodes based on the traffic network information, weight distribution of the multiple edges based on the real-time location of the rescue teams to determine multiple travel time weights.

[0067] Step S350: association and integration of the multiple nodes, the multiple edges, and the multiple travel time weights to construct the resource scheduling network graph.

[0068] Specifically, accident scene parameters related to the accident are retrieved, including chemical type, occurrence location, weather conditions, and surrounding environment, and key information is vectorized and encoded to form a state space representing the accident environment state. At the same time, according to the accident identification results in step S200, a set of pre-set emergency response actions (such as blocking, evacuation, cooling, fire extinguishing, chemical neutralization, etc.) are matched from the emergency action library to construct a set of optional response measures, and an action space is defined.

[0069] Based on the established state space and action space, a reinforcement learning model (such as DQN, PPO, etc.) is constructed to train a set of optimal action sequences for the current state with the goal of minimizing the response time and optimizing resource allocation, thereby formulating an emergency response scheme. For example, the model can evaluate the expected returns of different response action combinations in terms of reducing the spread range and shortening the response time, and select the scheme with the maximum return for output.

[0070] After executing the emergency response scheme, spatial resource information supporting the decision needs to be obtained, and rescue resource distribution data (such as fire station, chemical warehouse, medical team location), traffic network information (such as road smoothness, construction section), and real-time location of rescue teams are extracted in real time through platform interfaces as the basis for subsequent scheduling graph modeling.

[0071] According to the above resource data, first, the positions of various key facilities and teams are taken as multiple nodes in the graph. Then, multiple edges (representing road travel paths) are established between the nodes in combination with the traffic network information. Further, the edges are assigned weight values according to the real-time positions of the rescue teams and the travel states, and the weights can represent travel time consumption, traffic risk, or priority. For example, an edge represents a route from a warehouse to an accident point, and the weight of the edge can be determined by the current travel speed and the degree of rescue urgency.

[0072] The node set, edge set, and travel time weight constructed as described above are modeled together to generate a resource scheduling network graph. The network graph can be used for path planning, optimal scheduling of response resources, and subsequent dynamic monitoring. The graph structure has high scalability and can be dynamically adjusted as new resources or road states are added, supporting scheduling optimization and risk prediction on the graph structure using technologies such as graph neural networks.

[0073] Through the implementation of the above steps, an emergency strategy generation mechanism based on state-action pairs and centered on reinforcement learning can be quickly constructed according to the actual accident situation, and a multi-dimensional resource scheduling preparation can be completed with the aid of a spatial graph network, providing a basis for the next stage of dynamic monitoring and intelligent decision-making of the accident.

[0074] Further, step S400 includes:

[0075] Step S410: Dynamic analysis is performed in combination with the resource scheduling network graph and the multi-source synchronous data to obtain monitoring results, and the monitoring results include multiple monitoring indicators.

[0076] Step S420: When there are monitoring indicators that exceed a preset threshold among the multiple monitoring indicators, a hierarchical early warning signal is triggered, and real-time decision instructions are generated.

[0077] Step S430: Based on the real-time decision instructions, resource demand fields and evacuation path information are obtained.

[0078] Step S440: The resource demand fields are matched with the emergency response scheme to determine a list of material reserves.

[0079] Step S450: According to the evacuation path in combination with the list of material reserves, the priority of rescue is adjusted, an emergency execution instruction set is constructed, and the emergency execution instruction set is pushed to the emergency command center for emergency response to the chemical accident.

[0080] Specifically, the constructed resource scheduling network graph is fused in real time with the multi-source synchronous data obtained in the foregoing step S100, and through monitoring video, sensors, communication and other information within the coverage of the nodes, a plurality of key monitoring indicators such as toxic gas concentration, temperature change rate, personnel density and key equipment operating state are extracted. Through a graph neural network or a dynamic graph aggregation mechanism, the state of each node is dynamically updated to form a monitoring result set reflecting the accident situation.

[0081] A hierarchical threshold is set for different types of indicators (for example, a temperature greater than 80°C is a first-level early warning, and a temperature greater than 100°C is a second-level early warning), and the monitoring results are continuously subjected to threshold judgment. The plurality of indicators in the monitoring results are compared in real time with the corresponding preset threshold, and if any monitoring indicator exceeds the preset threshold, a hierarchical early warning signal (such as a blue, yellow, red three-level signal) corresponding to the indicator level is triggered immediately. When multiple indicators exceed the threshold, the corresponding level of early warning signal is triggered according to the highest priority principle. At the same time, according to the early warning level and the current resource graph state, a rule decision engine is called to generate real-time decision instructions, which are output in a structured form and cover response actions, resource calling suggestions, path adjustment, etc.

[0082] The content in the decision instructions is structured and parsed to extract key execution information, including resource demand fields (such as the type, quantity and professional team category of the equipment to be called) and evacuation path information (a set of optimal evacuation paths dynamically calculated based on the current traffic state). The parsing process calls a natural language parsing engine and a graph path search algorithm (such as Dijkstra) to complete.

[0083] According to the extracted resource demand field, it is matched and analyzed with the emergency response scheme prepared in the foregoing step S320 to determine whether all the demands are covered in the current response strategy. If there is a gap, a material reserve list is automatically supplemented, which is optimized in combination with the available location of the materials, the time efficiency and other factors.

[0084] Further combining the evacuation path information and the material reserve list, the priority order of the original rescue task is dynamically adjusted, such as preferentially placing hazardous material handling equipment in some passages and increasing the evacuation guide force at specific nodes. Finally, a set of multi-dimensional scheduling and coordination emergency execution instructions is constructed, covering key fields such as response content, task division, material scheduling and personnel path, and is pushed in real time to the emergency command center through an emergency platform interface to realize precise and efficient chemical accident emergency response closed loop.

[0085] Further, the step S410 includes:

[0086] Step S411: Spatially mapping the plurality of nodes in the resource scheduling network graph with the collection points of the multi-source synchronous data to establish a node-data association index table.

[0087] Step S412: continuously extracting real-time sensor data of the sensors, associating the real-time sensor data with the plurality of nodes in the resource scheduling network graph, and determining a plurality of node state attributes.

[0088] Step S413: identifying the smoke diffusion trend of the accident point, and cooperatively calculating in combination with the plurality of edges in the resource scheduling network graph to obtain a plurality of edge state attributes.

[0089] Step S414: when a new risk point appears, the plurality of travel time weights in the resource scheduling network graph are updated, and a plurality of weight coefficients are determined.

[0090] Step S415: based on the plurality of node state attributes, the plurality of edge state attributes, and the plurality of weight coefficients, the plurality of monitoring indicators are integrated according to the node-data association index table, and the plurality of monitoring indicators are added to the monitoring result.

[0091] Specifically, the GIS system is used for coordinate projection matching, the plurality of nodes (such as warehouses and intersections) in the resource scheduling network graph are spatially bound to the collection points (such as sensor deployment points and monitoring camera positions) of the multi-source synchronous data, a mapping relationship is established according to the coordinates of each node in the geographical space and the spatial distribution of the collection points, a node-data association index table is formed, and is used for subsequent fast binding and retrieval of monitoring data.

[0092] Real-time sensor data is continuously extracted from various sensors (such as temperature and humidity sensors, smoke detectors, and toxic gas concentration probes), and the real-time data is associated and mapped with the corresponding nodes in the resource scheduling network graph, so as to dynamically assign corresponding state attributes to each node, such as danger level, temperature distribution, and gas concentration. For example, according to the chlorine concentration data, the state attribute of the liquid chlorine storage tank node is determined to be “normal” (concentration below the safety threshold), “warning” (concentration close to the safety threshold), or “alarm” (concentration exceeding the safety threshold).

[0093] Based on the video monitoring system or the air quality detection equipment, the diffusion trend of the smoke or harmful gas of the chemical accident point is identified, and the diffusion path cooperatively calculated in combination with the plurality of edges (representing path connections) in the resource scheduling network graph, so as to obtain the state attributes of each edge, such as the risk level of passing through, the blocking state, and the crossing time.

[0094] When a new risk point (such as a sudden leakage area, a fire source, or a secondary explosion point) is detected through real-time analysis, the travel time weight in the resource scheduling network graph is immediately updated. The updating method includes adjusting the weight coefficient based on the risk level to determine the passability and the emergency degree of each path, so that the path planning avoids high-risk areas.

[0095] Based on the node-data association index table, the status attributes of each node, the status attributes of multiple edges, and the multiple weight coefficients after the passage time are integrated to finally construct a set of monitoring indicators that includes multiple dimensions such as concentration level, temperature index, passage time, and personnel distribution density. This set is then recorded in the current monitoring results to support the generation of the next early warning decision.

[0096] Through the above implementation steps, the resource scheduling network diagram and multi-source synchronous data can be accurately integrated to reflect the development trend of the accident in real time and generate multi-dimensional and high-precision monitoring indicators, which can effectively support the formulation of subsequent graded early warning and emergency response instructions, and significantly improve the perception capability and intelligent response level of chemical accidents.

[0097] In summary, the real-time monitoring-based emergency response method for chemical accidents provided in this application has the following beneficial effects:

[0098] This application's embodiments achieve accurate identification and rapid judgment of chemical accidents through real-time acquisition and fusion of multi-source heterogeneous data, combined with multimodal deep learning technology. Simultaneously, a resource scheduling network diagram constructed based on accident scenario and resource scheduling analysis optimizes resource allocation and scheduling efficiency in emergency response. Finally, through real-time monitoring, early warning, and rapid push of decision-making instructions, a closed-loop emergency response mechanism is formed, enabling continuous linkage and intelligent adjustment capabilities in emergency response. Overall, this method effectively improves the identification accuracy, response speed, and collaborative handling capabilities of chemical accidents, providing technical support and intelligent assurance for emergency management in high-risk scenarios.

[0099] Example 2, as Figure 2 As shown, based on the same inventive concept as in Embodiment 1 above, this application provides a real-time monitoring-based emergency response system for chemical accidents, the system comprising:

[0100] The data acquisition module 10 is used to acquire multi-source heterogeneous data in real time, perform time alignment based on the multi-source heterogeneous data, and generate multi-source synchronized data.

[0101] The accident identification module 20 is used to perform multimodal deep learning based on the multi-source synchronous data, extract multimodal features, identify chemical accidents according to the multimodal features, and obtain accident identification results.

[0102] The response analysis module 30 is used to retrieve accident scene parameters, perform accident response analysis based on the accident identification results, formulate an emergency response plan, execute the emergency response plan to perform resource scheduling analysis, and construct a resource scheduling network diagram.

[0103] An emergency response module 40 is configured to perform accident monitoring according to the resource scheduling network diagram in combination with the multi-source synchronous data, trigger a warning signal according to a monitoring result, generate a real-time decision instruction, and push the real-time decision instruction in combination with the emergency response scheme to an emergency command center for emergency response to the chemical accident.

[0104] Further, the data collection module 10 is further configured to perform the following steps:

[0105] The multi-source heterogeneous data is obtained through real-time collection by the multi-source data interface; the time window length is dynamically set to construct a time window, the multi-source heterogeneous data is aligned according to timestamps by using the time window to generate an aligned data set; timestamp offset analysis is performed according to the aligned data set to generate a plurality of offsets; it is determined whether the plurality of offsets exceeds the time window length, if any offset in the plurality of offsets exceeds the time window length, a data compensation instruction is triggered; the aligned data set is compensated for missing data by using the data compensation instruction to obtain the multi-source heterogeneous data.

[0106] Further, the data collection module 10 is further configured to perform the following steps:

[0107] The historical transmission delay statistical values of a plurality of data sources in the multi-source data interface are called, and the reference length of the time window is initialized according to the historical transmission delay statistical values; the transmission jitter rate of the plurality of data sources is monitored in real time, and an expansion parameter is generated when the transmission jitter rate exceeds a preset threshold; the reference length of the time window is dynamically expanded according to the expansion parameter to set the time window length.

[0108] Further, the accident identification module 20 is further configured to perform the following steps:

[0109] The multi-source synchronous data set is input into a multi-modal deep learning network for multi-modal analysis to obtain multi-modal data, the multi-modal data including video stream data, sensor time series signals, and structured text data; three-dimensional convolution analysis is performed based on the video stream data to extract spatio-temporal features; bidirectional analysis is performed based on the sensor time series signals to extract time series features; natural language processing is performed based on the structured text data to extract semantic keyword features; the spatio-temporal features, the time series features, and the semantic keyword features are weighted and fused to construct the multi-modal features.

[0110] Further, the response analysis module 30 is further configured to perform the following steps:

[0111] vectorization based on the accident scene parameters, defining a state space, matching emergency actions based on the accident identification results, defining an action space; reinforcement learning of the state space combined with the action space to formulate an emergency response scheme; performing resource scheduling analysis according to the emergency response scheme to obtain rescue resource distribution data, traffic network information, and real-time location of rescue teams; constructing multiple nodes based on the rescue resource distribution data, constructing multiple edges of the nodes based on the traffic network information, and assigning weights to the multiple edges based on the real-time location of the rescue teams to determine multiple travel time weights; associating and integrating the multiple nodes, the multiple edges, and the multiple travel time weights to construct the resource scheduling network graph.

[0112] Further, the emergency response module 40 of the embodiment of the application is further used to perform the following steps:

[0113] Performing dynamic analysis on the resource scheduling network graph and the multi-source synchronous data to obtain monitoring results, the monitoring results including multiple monitoring indicators; triggering a hierarchical early warning signal and generating real-time decision instructions when there are monitoring indicators exceeding a preset threshold among the multiple monitoring indicators; performing analysis based on the real-time decision instructions to obtain a resource demand field and evacuation path information; matching the resource demand field with the emergency response scheme to determine a material reserve list; adjusting rescue priorities according to the evacuation path combined with the material reserve list to construct an emergency execution instruction set, and pushing the emergency execution instruction set to the emergency command center for emergency response to the chemical accident.

[0114] Further, the emergency response module 40 of the embodiment of the application is further used to perform the following steps:

[0115] Performing spatial mapping of the multiple nodes in the resource scheduling network graph and the collection points of the multi-source synchronous data to establish a node-data association index table; continuously extracting real-time sensor data, associating the real-time sensor data with the multiple nodes in the resource scheduling network graph to determine multiple node state attributes; identifying smoke diffusion trends of accident points, and performing collaborative calculation combined with the multiple edges in the resource scheduling network graph to obtain multiple edge state attributes; when there are new risk points, performing weight updating of the multiple travel time weights in the resource scheduling network graph to determine multiple weight coefficients; integrating the multiple node state attributes, the multiple edge state attributes, and the multiple weight coefficients according to the node-data association index table to construct the multiple monitoring indicators, and adding the multiple monitoring indicators to the monitoring results.

[0116] The foregoing detailed description of the method for emergency response to chemical accidents under real-time monitoring enables those skilled in the art to clearly understand the system for emergency response to chemical accidents under real-time monitoring in the embodiments. For the system disclosed in Embodiment Two, since it corresponds to the method disclosed in Embodiment One, it has corresponding functional modules and beneficial effects. For the related parts, refer to the method part description.

[0117] The above description of disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for emergency response to a chemical incident under real-time monitoring, characterized in that, The method comprises: Real-time acquisition of multi-source heterogeneous data, time alignment based on the multi-source heterogeneous data, generation of multi-source synchronous data; According to the multi-source synchronous data, multi-modal deep learning is carried out, multi-modal features are extracted, chemical accident identification is carried out according to the multi-modal features, and accident identification results are obtained; Retrieve the accident scene parameters, analyze the accident response based on the accident identification results, develop an emergency response plan, perform resource scheduling analysis on the emergency response plan, and construct a resource scheduling network graph; According to the resource scheduling network graph combined with the multi-source synchronous data, the accident monitoring is carried out, the warning signal is triggered according to the monitoring result, the real-time decision instruction is generated, and the real-time decision instruction combined with the emergency response plan is pushed to the emergency command center to respond to the chemical accident; Retrieve the accident scene parameters, combine the accident identification results to analyze the accident response, develop an emergency response plan, perform resource scheduling analysis on the emergency response plan, and construct a resource scheduling network graph, the method comprising: Based on the accident scene parameters, vectorization is carried out, the state space is defined, the emergency action matching is carried out based on the accident identification results, and the action space is defined; The state space is combined with the action space for reinforcement learning to develop an emergency response plan; According to the emergency response plan, resource scheduling analysis is performed to obtain rescue resource distribution data, traffic network information and real-time location of rescue team; Based on the rescue resource distribution data, a plurality of nodes are constructed, based on the traffic network information, a plurality of edges of the plurality of nodes are constructed, based on the real-time location of the rescue team, a plurality of traffic time weights are distributed, and a plurality of traffic time weights are determined; The plurality of nodes, the plurality of edges and the plurality of traffic time weights are associated and integrated to construct the resource scheduling network graph; According to the resource scheduling network graph combined with multi-source synchronous data, the accident monitoring is carried out, the warning signal is triggered according to the monitoring result, the real-time decision instruction is generated, and the real-time decision instruction combined with the emergency response plan is pushed to the emergency command center to respond to the chemical accident, the method comprising: Dynamic analysis is carried out combined with the resource scheduling network graph and the multi-source synchronous data to obtain a monitoring result, and the monitoring result includes a plurality of monitoring indexes; When the plurality of monitoring indexes exist, the monitoring indexes exceeding the preset threshold are triggered, the real-time decision instruction is generated; Based on the real-time decision instruction, resource demand fields and evacuation path information are obtained; The resource demand field is matched with the emergency response plan to determine the material reserve list; According to the evacuation path combined with the material reserve list, the rescue priority is adjusted, the emergency execution instruction set is constructed, and the emergency execution instruction set is pushed to the emergency command center to respond to the chemical accident; Dynamic analysis is carried out combined with the resource scheduling network graph and the multi-source synchronous data to obtain a monitoring result, the method comprising: The plurality of nodes in the resource scheduling network graph are spatially mapped with the acquisition points of the multi-source synchronous data, and a node-data association index table is established; continuously extracting real-time sensing data of the sensors, associating the real-time sensing data with a plurality of nodes in the resource scheduling network graph, determining a plurality of node state attributes; identifying an accident point smoke diffusion trend, cooperatively calculating in combination with the plurality of edges in the resource scheduling network graph, obtaining a plurality of edge state attributes; when a new risk point occurs, then the plurality of travel time weights in the resource scheduling network graph are updated, a plurality of weight coefficients are determined; based on the plurality of node state attributes, the plurality of edge state attributes, and the plurality of weight coefficients, integrating according to the node-data association index table, constructing the plurality of monitoring indicators, and adding the plurality of monitoring indicators to the monitoring result.

2. The chemical incident emergency response method under real-time monitoring according to claim 1, characterized in that, real-time collection of multi-source heterogeneous data, time alignment based on the multi-source heterogeneous data, generation of multi-source synchronous data, the method comprising: real-time collection through a multi-source data interface to obtain multi-source heterogeneous data; dynamically setting a time window length to construct a time window, and aligning the multi-source heterogeneous data according to timestamps using the time window to generate an aligned data set; timestamp offset analysis according to the aligned data set to generate a plurality of offsets; determining whether the plurality of offsets exceeds the time window length, if any offset in the plurality of offsets exceeds the time window length, triggering a data compensation instruction; missing compensation of the aligned data set through the data compensation instruction to obtain the multi-source heterogeneous data.

3. The chemical incident emergency response method under real-time monitoring according to claim 2, wherein, The process of dynamically setting the time window length comprises: retrieve historical transmission delay statistics of a plurality of data sources in the multi-source data interface, and initialize the reference length of the time window according to the historical transmission delay statistics; real-time monitoring of the transmission jitter rate of the plurality of data sources, generating an expansion parameter when the transmission jitter rate exceeds a preset threshold; dynamically expanding the reference length of the time window according to the expansion parameter to set the time window length.

4. The chemical incident emergency response method under real-time monitoring according to claim 1, wherein, Multi-modal deep learning based on the multi-source synchronous data, extracting multi-modal features, the method comprising: inputting the multi-source synchronous data set into a multi-modal deep learning network for multi-modal analysis to obtain multi-modal data, the multi-modal data including video stream data, sensor time series signals, and structured text data; three-dimensional convolution analysis based on the video stream data to extract spatio-temporal features; bidirectional analysis based on the sensor time series signals to extract time series features; natural language processing based on the structured text data to extract semantic keyword features; weighted fusion of the spatio-temporal features, the time series features, and the semantic keyword features to construct the multi-modal features.

5. A chemical incident emergency response system under real-time monitoring, characterized in that, The system is used to execute the real-time monitoring based chemical accident emergency response method of any one of claims 1-4, comprising: a data acquisition module for real-time collection of multi-source heterogeneous data, time alignment based on the multi-source heterogeneous data, and generation of multi-source synchronous data; an accident identification module for multi-modal deep learning based on the multi-source synchronous data, extraction of multi-modal features, chemical accident identification according to the multi-modal features, and obtaining of an accident identification result; The response analysis module is configured to call accident scene parameters, perform accident response analysis based on the accident identification result, formulate an emergency response scheme, perform resource scheduling analysis according to the emergency response scheme, and construct a resource scheduling network diagram. The emergency response module is configured to perform accident monitoring according to the resource scheduling network diagram in combination with the multi-source synchronous data, trigger a warning signal according to a monitoring result, generate a real-time decision instruction, and push the real-time decision instruction to an emergency command center in combination with the emergency response scheme to perform emergency response to the chemical accident.

Images